Skip to main content

Voice Agents overview

Voice agents hold real-time spoken conversations. They run over LiveKit for low-latency audio and can handle both inbound and outbound calls.

Call types​

  • Inbound — the agent answers calls placed to a connected number.
  • Outbound — the agent places calls (e.g. reminders, follow-ups).

Real-time providers​

Voice agents use a real-time speech model from one of:

  • Gemini
  • OpenAI

For agents that pair a real-time voice layer with a separate reasoning model, you also choose an LLM model.

Lifecycle​

  1. Create the agent — pick a voice, provider, and (where applicable) an LLM.
  2. Test it in the in-browser LiveKit console.
  3. Attach a phone number via telephony.

Voice usage draws credits based on the chosen model — see Credits & Billing.